many observers - Niall Cardin

Recommendations

Info

philosopher explains the Self-Indication Assumption: (See e.g. Bostrom 2002; Bostrom 2003, http://www.anthropic-principle.com/preprints/olum/sia.pdf) Having considered this I wondered whether by setting things up a little differently from the motivating example I could get a different conclusion, I found that I could - and I think it highlights some of the implicit assumptions being made: Better defined experiment, to promote discussion We need to be slightly more powerful beings for this - capable of creating life at will. I guess we could do this using combinations of women and men (and possibly alcohol), if we were prepared to wait... Anyway, practicalities aside - we decide to run a lab experiment (possibly repeated <strong>many</strong> times). We will generate beings according to the flips of two coins. The beings will all know and understand all details of the experiment, which are • We have two rooms, as big as we need. To avoid mutiny each room is split into hermetically sealed (but life supporting) units, so that inhabitants have absolutely no idea how <strong>many</strong> others there might be around them. • We have two fair coins, Coin A (C A ), and Coin B (C B ), with independent (random) outcomes. • C A is flipped: – If C A = heads we fill room A with N A individuals. In particular, think about the case N A = 1 – If C A = tails we don’t put anyone in room A. • C B is flipped: – If C B = heads we fill room B with N B individuals. In particular, think about the case N B >> 1 – If C B = tails we don’t put anyone in room B. • If there are any subjects to interview, we interview each of them - asking them what probability they assign the event “I am in room A” So, as a participant, created in the course of the experiment, what should I say First: what are the events that we want to think about, and what are events that we know how to define, and are clear on Want to think about • The probability that I am in room A • The probability that either of the coins landed heads given that I am here to ask the question 2
Questions we can define well • The probability that the coins come down heads or tails, without conditioning on anyone in the rooms to wonder which room they’re in. • The probability that there will be some people in rooms A, B respectively So we know how to reason from the perspective of the experimenter, but not from the perspective of the subjects. This is probably what Bostrom means by “Indexical vs. Non-Indexical” reasoning. Does it help to think about bets that one should accept or not Suppose I am asked whether I will accept a bet about whether or not I’m in room A. Let’s imagine that N B >> N A , and I’m offered 4 : 1 odds. Well, depending on whether I should think like Bostrom or prefer to use “Proof by symbolic manipulation” (below) I appear to get different answers. First though we should be more clear about the set-up: As a participant in this study I get to know all the details of questions asked of me, and hence I know exactly the structure of this betting procedure. There are multiple possibilities: 1. Every participant is asked whether or not they want to participate in the bet 2. Only one person from each room is asked whether they want to participate in the bet 3. The experimenter only asks people in room B (he wants to make some cash, sneaky devil) and so on. Obviously in case 3 we should not accept the bet. In case 2 we do want to accept the bet, indeed conditional on being asked we’re now evenly split between thinking that we’re in room A or room B. So, what about case 1 The following argument says that we should not accept the bet, hence that we’re actually pretty likely to be in room B. Balance argument If the experimenter is going to ask everyone to bet (and let’s assume for now that they all say yes), then when coin C B = heads, the experimenter is going to make $N B . When C B = tails then the experimenter will lose either $0 (C A is also tails) or $4N A (C A is heads). Overall the experimenters expectation, E e is: E e = 1 4 (N B − 4N A + 0 + (N B − 4N A )) = 1 2 (N B − 4N A ) (3) From this we see that the experimenter gets a positive expectation from these bets whenever N B > 4N A . Note that we could replace this 4 with any n, and in order for the experimenter to profit he would simply need to ensure that N B > nN A . 3
Page 1: Inference using the potential popul

many observers - Niall Cardin

Create successful ePaper yourself

Delete template?

Save as template?